File size: 10,640 Bytes
f268b80
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
# origin: https://github.com/intel/openvino-ai-plugins-gimp/blob/ae93e7291fab6d372c958da18e497acb9d927055/gimpopenvino/tools/openvino_common/models_ov/stable_diffusion_engine.py#L748

import os
from typing import Union, Optional, Any, List, Dict

import torch
from openvino.runtime import Core
from diffusers import DiffusionPipeline, LCMScheduler, ImagePipelineOutput
from diffusers.image_processor import VaeImageProcessor
from transformers import CLIPTokenizer


class LatentConsistencyEngine(DiffusionPipeline):
    def __init__(
        self,
            model="SimianLuo/LCM_Dreamshaper_v7",
            tokenizer="openai/clip-vit-large-patch14",
            device=["CPU", "CPU", "CPU"],
    ):
        super().__init__()
        try:
            self.tokenizer = CLIPTokenizer.from_pretrained(model, local_files_only=True)
        except:
            self.tokenizer = CLIPTokenizer.from_pretrained(tokenizer)
            self.tokenizer.save_pretrained(model)

        self.core = Core()
        self.core.set_property({'CACHE_DIR': os.path.join(model, 'cache')})  # adding caching to reduce init time
        # text features

        print("Text Device:", device[0])
        self.text_encoder = self.core.compile_model(os.path.join(model, "text_encoder.xml"), device[0])
        self._text_encoder_output = self.text_encoder.output(0)

        # diffusion
        print("unet Device:", device[1])
        self.unet = self.core.compile_model(os.path.join(model, "unet.xml"), device[1])
        self._unet_output = self.unet.output(0)
        self.infer_request = self.unet.create_infer_request()

        # decoder
        print("Vae Device:", device[2])

        self.vae_decoder = self.core.compile_model(os.path.join(model, "vae_decoder.xml"), device[2])
        self.infer_request_vae = self.vae_decoder.create_infer_request()
        self.safety_checker = None #pipe.safety_checker
        self.feature_extractor = None #pipe.feature_extractor
        self.vae_scale_factor = 2 ** 3
        self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
        self.scheduler =  LCMScheduler(
            beta_start=0.00085,
            beta_end=0.012,
            beta_schedule="scaled_linear"
        )

    def _encode_prompt(
        self,
        prompt,
        num_images_per_prompt,
        prompt_embeds: None,
    ):
        r"""
        Encodes the prompt into text encoder hidden states.
        Args:
            prompt (`str` or `List[str]`, *optional*):
                prompt to be encoded
            num_images_per_prompt (`int`):
                number of images that should be generated per prompt
            prompt_embeds (`torch.FloatTensor`, *optional*):
                Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not
                provided, text embeddings will be generated from `prompt` input argument.
        """

        if prompt_embeds is None:

            text_inputs = self.tokenizer(
                prompt,
                padding="max_length",
                max_length=self.tokenizer.model_max_length,
                truncation=True,
                return_tensors="pt",
            )
            text_input_ids = text_inputs.input_ids
            untruncated_ids = self.tokenizer(
                prompt, padding="longest", return_tensors="pt"
            ).input_ids

            if untruncated_ids.shape[-1] >= text_input_ids.shape[
                -1
            ] and not torch.equal(text_input_ids, untruncated_ids):
                removed_text = self.tokenizer.batch_decode(
                    untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
                )

            prompt_embeds = self.text_encoder(text_input_ids, share_inputs=True, share_outputs=True)
            prompt_embeds = torch.from_numpy(prompt_embeds[0])

        bs_embed, seq_len, _ = prompt_embeds.shape
        # duplicate text embeddings for each generation per prompt
        prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
        prompt_embeds = prompt_embeds.view(
            bs_embed * num_images_per_prompt, seq_len, -1
        )

        # Don't need to get uncond prompt embedding because of LCM Guided Distillation
        return prompt_embeds

    def run_safety_checker(self, image, dtype):
        if self.safety_checker is None:
            has_nsfw_concept = None
        else:
            if torch.is_tensor(image):
                feature_extractor_input = self.image_processor.postprocess(
                    image, output_type="pil"
                )
            else:
                feature_extractor_input = self.image_processor.numpy_to_pil(image)
            safety_checker_input = self.feature_extractor(
                feature_extractor_input, return_tensors="pt"
            )
            image, has_nsfw_concept = self.safety_checker(
                images=image, clip_input=safety_checker_input.pixel_values.to(dtype)
            )
        return image, has_nsfw_concept

    def prepare_latents(
        self, batch_size, num_channels_latents, height, width, dtype, latents=None
    ):
        shape = (
            batch_size,
            num_channels_latents,
            height // self.vae_scale_factor,
            width // self.vae_scale_factor,
        )
        if latents is None:
            latents = torch.randn(shape, dtype=dtype)
        # scale the initial noise by the standard deviation required by the scheduler
        return latents

    def get_w_embedding(self, w, embedding_dim=512, dtype=torch.float32):
        """
        see https://github.com/google-research/vdm/blob/dc27b98a554f65cdc654b800da5aa1846545d41b/model_vdm.py#L298
        Args:
        timesteps: torch.Tensor: generate embedding vectors at these timesteps
        embedding_dim: int: dimension of the embeddings to generate
        dtype: data type of the generated embeddings
        Returns:
        embedding vectors with shape `(len(timesteps), embedding_dim)`
        """
        assert len(w.shape) == 1
        w = w * 1000.0

        half_dim = embedding_dim // 2
        emb = torch.log(torch.tensor(10000.0)) / (half_dim - 1)
        emb = torch.exp(torch.arange(half_dim, dtype=dtype) * -emb)
        emb = w.to(dtype)[:, None] * emb[None, :]
        emb = torch.cat([torch.sin(emb), torch.cos(emb)], dim=1)
        if embedding_dim % 2 == 1:  # zero pad
            emb = torch.nn.functional.pad(emb, (0, 1))
        assert emb.shape == (w.shape[0], embedding_dim)
        return emb

    @torch.no_grad()
    def __call__(
        self,
        prompt: Union[str, List[str]] = None,
        height: Optional[int] = 512,
        width: Optional[int] = 512,
        guidance_scale: float = 7.5,
        scheduler = None,
        num_images_per_prompt: Optional[int] = 1,
        latents: Optional[torch.FloatTensor] = None,
        num_inference_steps: int = 4,
        lcm_origin_steps: int = 50,
        prompt_embeds: Optional[torch.FloatTensor] = None,
        output_type: Optional[str] = "pil",
        return_dict: bool = True,
        model: Optional[Dict[str, any]] = None,
        seed: Optional[int] = 1234567,
        cross_attention_kwargs: Optional[Dict[str, Any]] = None,
        callback = None,
        callback_userdata = None
    ):

        # 1. Define call parameters
        if prompt is not None and isinstance(prompt, str):
            batch_size = 1
        elif prompt is not None and isinstance(prompt, list):
            batch_size = len(prompt)
        else:
            batch_size = prompt_embeds.shape[0]

        if seed is not None:
            torch.manual_seed(seed)

        #print("After Step 1: batch size is ", batch_size)
        # do_classifier_free_guidance = guidance_scale > 0.0
        # In LCM Implementation:  cfg_noise = noise_cond + cfg_scale * (noise_cond - noise_uncond) , (cfg_scale > 0.0 using CFG)

        # 2. Encode input prompt
        prompt_embeds = self._encode_prompt(
            prompt,
            num_images_per_prompt,
            prompt_embeds=prompt_embeds,
        )
        #print("After Step 2: prompt embeds is ", prompt_embeds)
        #print("After Step 2: scheduler is ", scheduler )
        # 3. Prepare timesteps
        self.scheduler.set_timesteps(num_inference_steps, original_inference_steps=lcm_origin_steps)
        timesteps = self.scheduler.timesteps

        #print("After Step 3: timesteps is ", timesteps)

        # 4. Prepare latent variable
        num_channels_latents = 4
        latents = self.prepare_latents(
            batch_size * num_images_per_prompt,
            num_channels_latents,
            height,
            width,
            prompt_embeds.dtype,
            latents,
        )
        latents = latents * self.scheduler.init_noise_sigma

        #print("After Step 4: ")
        bs = batch_size * num_images_per_prompt

        # 5. Get Guidance Scale Embedding
        w = torch.tensor(guidance_scale).repeat(bs)
        w_embedding = self.get_w_embedding(w, embedding_dim=256)
        #print("After Step 5: ")
        # 6. LCM MultiStep Sampling Loop:
        with self.progress_bar(total=num_inference_steps) as progress_bar:
            for i, t in enumerate(timesteps):
                if callback:
                    callback(i+1, callback_userdata)

                ts = torch.full((bs,), t, dtype=torch.long)

                # model prediction (v-prediction, eps, x)
                model_pred = self.unet([latents, ts, prompt_embeds, w_embedding],share_inputs=True, share_outputs=True)[0]

                # compute the previous noisy sample x_t -> x_t-1
                latents, denoised = self.scheduler.step(
                    torch.from_numpy(model_pred), t, latents, return_dict=False
                )
                progress_bar.update()

        #print("After Step 6: ")

        #vae_start = time.time()

        if not output_type == "latent":
            image = torch.from_numpy(self.vae_decoder(denoised / 0.18215, share_inputs=True, share_outputs=True)[0])
        else:
            image = denoised

        #print("vae decoder done", time.time() - vae_start)
        #post_start = time.time()

        #if has_nsfw_concept is None:
        do_denormalize = [True] * image.shape[0]
        #else:
        #    do_denormalize = [not has_nsfw for has_nsfw in has_nsfw_concept]

        #print ("After do_denormalize: image is ", image)

        image = self.image_processor.postprocess(
            image, output_type=output_type, do_denormalize=do_denormalize
        )

        return ImagePipelineOutput([image[0]])